Gradient-based model-predictive control for green urban mobility in traffic networks

Abstract

To deal with the traffic congestion and emissions, traffic-responsive
control approaches can be used. The main aim of the control is then to
use the existing capacity of the network efficiently, and to reduce the
harmful economical and environmental effects of heavy traffic. In this
paper, we design a highly efficient model-predictive control system that
uses a gradient-based approach to solve the optimization problem, which
has been reformulated as a two-point boundary value problem. A
gradient-based approach computes the derivatives to find the optimal
value. Therefore, the optimization problem should involve only smooth
functions. Hence, for nonsmooth functions that may appear in the
internal model of the MPC controller, we propose smoothening approaches.
The controller then uses an integrated smooth flow and emission model,
where the control objective is to reduce a weighted combination of the
total time spent and total emissions of the vehicles. We perform
simulations to compare the efficiency and the CPU time of the smooth and
nonsmooth optimization approaches. The simulation results show that the
smooth approach significantly outperforms the nonsmooth one both in the
CPU time and in the optimal objective value.